Key points are not available for this paper at this time.
Recent advances in unmanned aerial vehicles (UAVs) technology and deep learning have revolutionized agricultural monitoring, yet challenges remain in processing low-resolution field imagery for precision floriculture. Here, we presented an innovative approach combining state-of-the-art super-resolution reconstruction (SRR) and object detection for accurate rose growth monitoring in large-scale greenhouse environments. We introduced MambaIR, a novel SRR algorithm based on selective state-space models, which significantly outperforms existing methods in enhancing low-resolution UAV imagery (PSNR: 28.34 dB, SSIM: 77.07 %). We also developed ROSE-YOLO, an improved object detection model tailored for rose identification, achieving 95.3 % mean average precision (mAP) on high-resolution images. The synergy between MambaIR and ROSE-YOLO enables 94.4 % mAP on reconstructed super-resolution images, nearly matching high-resolution performance. Through comprehensive experiments and Grad-CAM visualizations, we demonstrated our method's superior focus on key rose features and identify an optimal super-resolution magnification factor balancing detail enhancement and computational efficiency. This integrated approach overcomes resolution limitations in UAV-based agricultural monitoring, offering a scalable and accurate solution for rose growth assessment. Our method reduces technical barriers, offering a scalable and cost-effective solution for greenhouse monitoring by addressing low-resolution imagery challenges and enhancing decision-making processes. This research lays the groundwork for broader applications of UAV and AI technologies in sustainable agriculture. The findings pave the way for advanced, data-driven precision agriculture, integrating deep learning with remote sensing methodologies to improve floriculture management.
Zhao et al. (Wed,) studied this question.